Benchmark Almanac

SecRespond

Claude Opus 4.7 79% leads since July 2026
Not in model cards

Agents investigate compromised-host snapshots and write incident analyses and remediation plans. A model evaluator applies expert checklists separately to detection and response planning.

The ten ranges contain sanitized synthetic forensic evidence. The maker explicitly prohibits combining detection and planning into one score.

Made by
Alibaba NLP
Released
July 2026
Kind
Finding and exploiting real bugs
Level
Expert
Scored
Graded by a model
Answers
Agent task
Tasks
All public
Measures
Capability
Language
English
In and out
Text → Actions, Text
Interaction
Agentic
Environment
Terminal
Items from
Generated
Refresh
One fixed set
Run by
Anyone
In views
Agents
Strongest model tested
GPT-5.4 Pro #19 on the capability index
Last model added
June 2026
Scores refreshed
November 20th, 2026

In the authors’ words

Large Language Model (LLM) agents are increasingly adopted in real-world security operations with access to host artifacts and command-line interfaces (CLIs), making it critical to thoroughly assess their security capabilities. However, existing cybersecurity benchmarks focus on pre-compromise settings where agents are placed in a clean and idealized environment before an attack occurs. This leaves the post-compromise setting underexplored. To address this gap, we introduce SecRespond, the first benchmark for evaluating LLM agents on the post-compromise incident-response workflow. Given a forensic disk snapshot of a compromised host together with the alerts, vulnerability scans, and baseline checks reported by a host security product, agents are required to produce forensic reports on intrusions, baseline risks, and vulnerability risks, together with a remediation plan. We instantiate this task across 10 cyber ranges, each constructed from a distinct compromised cloud host, spanning 4 entry-point types, 21 ATT&CK techniques, and 5 operating systems. We evaluate 23 frontier LLMs on the OpenCode agent harness. Experimental results show that although current agents can reliably uncover the problems exposed by alerts, they struggle to proactively investigate the disk for silent intrusions and to produce comprehensive, verified remediation plans, with no model achieving complete detection and remediation on any single range. This reveals a fundamental bottleneck in building agents for real-world incident response. The benchmark is publicly available at https://github.com/Alibaba-NLP/qqr/tree/main/data/secrespond.

Abstract of SecRespond: Benchmarking AI Agents for Real-World Post-Compromise Incident Response (arXiv:2607.26791, CC0)
23
Models scored in its paper

Claude Opus 4.7 leads with 79% among models released by June 2026

Best score per model from the benchmark's paper, read on November 20th, 2026. A paper’s table doesn’t add newer models. The headline is the published overall detection percentage. The paper scores planning separately. Where a model ran with several settings or harnesses, we keep its best.

The newest model with a score here came out in June 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

  1. 1 Claude Opus 4.7 Nine ranges with responses 79%
  2. 2 Claude Opus 4.6 78.2%
  3. 3 Qwen3.7-Max 78%
  4. 4 GLM-5.1 76.3%
  5. 5 Qwen3.7-Plus 75.6%
  6. 6 Claude Sonnet 4.6 74.7%
  7. 7 DeepSeek-V4-Pro 73.2%
  8. 8 GPT-5.4 Pro 72.5%
  9. 9 GPT-5.5 70.7%
  10. 10 Claude Sonnet 4.5 69.7%
  11. 11 Claude Opus 4.5 69.5%
  12. 12 Qwen 3.6 Plus 68.4%
  13. 13 GPT-5.2 Pro 66.2%
  14. 14 Kimi K2.6 65.8%
  15. 15 GLM-5 64.9%
  16. 16 Qwen3.5-Plus 63.5%
  17. 17 Kimi K2.5 62.1%
  18. 18 GPT-5.4 60.7%
  19. 19 MiniMax-M2.7 60.1%
  20. 20 DeepSeek-V3.2 59.2%
  21. 21 Gemini 3 Flash 57.4%
  22. 22 Gemini 3.1 Pro 54.8%
  23. 23 MiniMax-M2.5 52.4%

Source: The SecRespond paper

#ModelScoreSettingDateSource
1 Claude Opus 4.7 79% Nine ranges with responses Apr 2026 Official leaderboard
2 Claude Opus 4.6 78.2% Feb 2026 Official leaderboard
3 Qwen3.7-Max 78% May 2026 Official leaderboard
4 GLM-5.1 76.3% Apr 2026 Official leaderboard
5 Qwen3.7-Plus 75.6% Jun 2026 Official leaderboard
6 Claude Sonnet 4.6 74.7% Feb 2026 Official leaderboard
7 DeepSeek-V4-Pro 73.2% Apr 2026 Official leaderboard
8 GPT-5.4 Pro 72.5% Mar 2026 Official leaderboard
9 GPT-5.5 70.7% Apr 2026 Official leaderboard
10 Claude Sonnet 4.5 69.7% Sep 2025 Official leaderboard
11 Claude Opus 4.5 69.5% Nov 2025 Official leaderboard
12 Qwen 3.6 Plus 68.4% Mar 2026 Official leaderboard
13 GPT-5.2 Pro 66.2% Dec 2025 Official leaderboard
14 Kimi K2.6 65.8% Apr 2026 Official leaderboard
15 GLM-5 64.9% Feb 2026 Official leaderboard
16 Qwen3.5-Plus 63.5% Feb 2026 Official leaderboard
17 Kimi K2.5 62.1% Jan 2026 Official leaderboard
18 GPT-5.4 60.7% Mar 2026 Official leaderboard
19 MiniMax-M2.7 60.1% Mar 2026 Official leaderboard
20 DeepSeek-V3.2 59.2% Dec 2025 Official leaderboard
21 Gemini 3 Flash 57.4% Dec 2025 Official leaderboard
22 Gemini 3.1 Pro 54.8% Feb 2026 Official leaderboard
23 MiniMax-M2.5 52.4% Feb 2026 Official leaderboard

Rankings by part

We read 21 parts from the benchmark’s paper, each ranking models on one part of the test, such as Overall planning, SSH-Miner detection and SSH-Miner planning. Claude Opus 4.7 leads 4 of them, and 8 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.

Overall planning

23 models we track have a score here.

  1. 1 Claude Opus 4.7 65.7%
  2. 2 Claude Sonnet 4.5 ↑8 61.2%
  3. 3 GLM-5.1 ↑1 59.2%
  4. 4 Claude Sonnet 4.6 ↑2 59.1%
  5. 5 Qwen3.7-Plus 58.8%
  6. 6 Claude Opus 4.6 ↓4 58%
  7. 7 Claude Opus 4.5 ↑4 56.4%
  8. 8 Qwen3.7-Max ↓5 54.3%
  9. 9 DeepSeek-V4-Pro ↓2 53.4%
  10. 10 Kimi K2.5 ↑7 50.7%

Number one over time

Lab claims count from the day the card came out, independent runs from the model’s release.

Anthropic
Benchmark paper Not tested since Jul 2026

Source: The SecRespond paper

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

Plot with other benchmarks
In the benchmark’s paper Best so far
0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) Claude Opus 4.7 Claude Opus 4.6 Claude Sonnet 4.5 0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) Claude Opus 4.7 Claude Sonnet 4.5

Source: Official SecRespond leaderboard

No lab model card reports it yet

We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.

More in Cybersecurity

All 138 in Cybersecurity
  • ExploitGym In use 869 real-vulnerability exploit-development tasks (userspace, V8, Linux kernel). Success means capturing a flag through the intended vulnerability. GPT-6 Astra 42.4%
  • ExploitBench In use 41 recent V8 JavaScript-engine N-day vulnerabilities. Progress measured by 16 grader-verified capability flags across five tiers. GPT-6 Astra 100%
  • ExploitBench - Internal Port In use Agents develop working exploits for recently disclosed vulnerabilities to limit contamination. Scored as arbitrary code execution success rate. GPT-6.2 Astra 53.5%
  • Atomic Challenges In use Agents solve 22 skill-level offensive security challenges in vulnerability research, network attack simulation and evasion. Scored as challenges solved. GPT-6 Astra 90.9%
  • MiMo Cyber Bench In use Agents do in-house cybersecurity tasks that add to public vulnerability reproduction and exploitation benchmarks. The report gives no task count or grading details. MiMo-V2.6 Pro 81.7%
  • PatchBench (OpenAI) In use Models patch security flaws. Automated tests and model-based review check that each patch fixes the flaw, keeps the intended behavior and is ready to merge. GPT-6.2 Astra 79.7%

Data audit

All 10 checks pass. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 4 pass 1 note
  • Pass Independent runs Scores from the benchmark’s paper (23 entries).
  • Pass Models scored 23 models we track.
  • Pass Newest score The newest is from June 2026, for Qwen3.7-Plus.
  • Note Frontier models None of the top 10 has a score. The strongest tested is GPT-5.4 Pro (#19).
  • Pass Headroom The best is 79%, under 85%, so it still separates models.
Leaderboard import 3 pass
  • Pass Importer We read the board on arxiv.org: 23 rows for models we track at the last read.
  • Pass Last read November 20th, 2026.
  • Pass Read worked The last read worked.
Consistency 1 pass 1 note
  • Note Lab claims No model has both a lab number and an independent run.
  • Pass Known issues No published audit found problems.

How we check